modified: phenotyping.py

auton_survival/phenotyping.py

@@ -24,6 +24,8 @@
 """Utilities to phenotype individuals based on similar survival
 characteristics."""
 
+from random import random
+from re import I
 import numpy as np
 import pandas as pd
 
@@ -32,6 +34,7 @@
 from sklearn import cluster, decomposition, mixture
 
 from auton_survival.utils import _get_method_kwargs
+from auton_survival.experiments import CounterfactualSurvivalRegressionCV
 
 
 class Phenotyper:
@@ -408,9 +411,64 @@ def fit_phenotype(self, features):
 
     return self.fit(features).phenotype(features)
 
-class CoxMixturePhenotyper:
+class SurvivalVirtualTwinsPhenotyper(object):
 
   """"Not Yet Implemented"""
 
-  def __init__(self):
-    raise NotImplementedError()
+
+  _VALID_PHENO_METHODS = ['rsf']
+
+  def __init__(self,
+               cf_method='dcph',
+               phenotyping_method='rsf', 
+               cf_method_hyperparams=None,
+               phenotyping_method_hyperparams=None,
+               random_seed=0):
+
+    raise NotImplementedError()
+
+    assert cf_method in CounterfactualSurvivalRegressionCV._VALID_CF_METHODS, "Invalid Counterfactual Method: "+cf_method
+    assert phenotyping_method in self._VALID_PHENO_METHODS, "Invalid Phenotyping Method: "+phenotyping_method
+
+    if cf_method_hyperparams is None:
+      cf_method_hyperparams = {}
+    if phenotyping_method_hyperparams is None:
+      phenotyping_method_hyperparams = {}
+
+    self.cf_method = cf_method
+    self.phenotyping_method = phenotyping_method
+    self.random_seed = random_seed
+
+  def fit(self, features, outcomes, interventions, horizon):
+
+    raise NotImplementedError()
+
+    cf_model = CounterfactualSurvivalRegressionCV(**self.cf_method_hyperparams)
+
+    self.cf_model = cf_model.fit(features, outcomes, interventions) 
+
+    times = np.unique(outcomes.times.values)
+    cf_predictions = self.cf_model.predict_counterfactual_survival(features, interventions, times)
+
+    ite_estimates = cf_predictions[1] - cf_predictions[0]
+
+    if self.phenotyping_method == 'rsf':
+
+      from sklearn.ensemble import RandomForestRegressor
+
+      pheno_model = RandomForestRegressor(**self.phenotyping_method_hyperparams)
+      pheno_model.fit(features.values, ite_estimates)
+
+    self.pheno_model = pheno_model
+
+  def predict(self, features):
+
+    raise NotImplementedError()
+
+    phenotype_preds=  self.pheno_model.predict(features)
+    phenotype_preds = (phenotype_preds -  phenotype_preds.min()) / (phenotype_preds.max() - phenotype_preds.min())
+    return phenotype_preds 
+
+
+
+